xref: /freebsd/sys/kern/uipc_socket.c (revision adfa0adec0b5d7c19c220a85ef6ca729235ed172)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * Copyright (c) 2004 The FreeBSD Foundation
5  * Copyright (c) 2004-2006 Robert N. M. Watson
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 4. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)uipc_socket.c	8.3 (Berkeley) 4/15/94
32  */
33 
34 /*
35  * Comments on the socket life cycle:
36  *
37  * soalloc() sets of socket layer state for a socket, called only by
38  * socreate() and sonewconn().  Socket layer private.
39  *
40  * sodealloc() tears down socket layer state for a socket, called only by
41  * sofree() and sonewconn().  Socket layer private.
42  *
43  * pru_attach() associates protocol layer state with an allocated socket;
44  * called only once, may fail, aborting socket allocation.  This is called
45  * from socreate() and sonewconn().  Socket layer private.
46  *
47  * pru_detach() disassociates protocol layer state from an attached socket,
48  * and will be called exactly once for sockets in which pru_attach() has
49  * been successfully called.  If pru_attach() returned an error,
50  * pru_detach() will not be called.  Socket layer private.
51  *
52  * pru_abort() and pru_close() notify the protocol layer that the last
53  * consumer of a socket is starting to tear down the socket, and that the
54  * protocol should terminate the connection.  Historically, pru_abort() also
55  * detached protocol state from the socket state, but this is no longer the
56  * case.
57  *
58  * socreate() creates a socket and attaches protocol state.  This is a public
59  * interface that may be used by socket layer consumers to create new
60  * sockets.
61  *
62  * sonewconn() creates a socket and attaches protocol state.  This is a
63  * public interface  that may be used by protocols to create new sockets when
64  * a new connection is received and will be available for accept() on a
65  * listen socket.
66  *
67  * soclose() destroys a socket after possibly waiting for it to disconnect.
68  * This is a public interface that socket consumers should use to close and
69  * release a socket when done with it.
70  *
71  * soabort() destroys a socket without waiting for it to disconnect (used
72  * only for incoming connections that are already partially or fully
73  * connected).  This is used internally by the socket layer when clearing
74  * listen socket queues (due to overflow or close on the listen socket), but
75  * is also a public interface protocols may use to abort connections in
76  * their incomplete listen queues should they no longer be required.  Sockets
77  * placed in completed connection listen queues should not be aborted for
78  * reasons described in the comment above the soclose() implementation.  This
79  * is not a general purpose close routine, and except in the specific
80  * circumstances described here, should not be used.
81  *
82  * sofree() will free a socket and its protocol state if all references on
83  * the socket have been released, and is the public interface to attempt to
84  * free a socket when a reference is removed.  This is a socket layer private
85  * interface.
86  *
87  * NOTE: In addition to socreate() and soclose(), which provide a single
88  * socket reference to the consumer to be managed as required, there are two
89  * calls to explicitly manage socket references, soref(), and sorele().
90  * Currently, these are generally required only when transitioning a socket
91  * from a listen queue to a file descriptor, in order to prevent garbage
92  * collection of the socket at an untimely moment.  For a number of reasons,
93  * these interfaces are not preferred, and should be avoided.
94  */
95 
96 #include <sys/cdefs.h>
97 __FBSDID("$FreeBSD$");
98 
99 #include "opt_inet.h"
100 #include "opt_mac.h"
101 #include "opt_zero.h"
102 #include "opt_compat.h"
103 
104 #include <sys/param.h>
105 #include <sys/systm.h>
106 #include <sys/fcntl.h>
107 #include <sys/limits.h>
108 #include <sys/lock.h>
109 #include <sys/mac.h>
110 #include <sys/malloc.h>
111 #include <sys/mbuf.h>
112 #include <sys/mutex.h>
113 #include <sys/domain.h>
114 #include <sys/file.h>			/* for struct knote */
115 #include <sys/kernel.h>
116 #include <sys/event.h>
117 #include <sys/eventhandler.h>
118 #include <sys/poll.h>
119 #include <sys/proc.h>
120 #include <sys/protosw.h>
121 #include <sys/socket.h>
122 #include <sys/socketvar.h>
123 #include <sys/resourcevar.h>
124 #include <sys/signalvar.h>
125 #include <sys/sysctl.h>
126 #include <sys/uio.h>
127 #include <sys/jail.h>
128 
129 #include <vm/uma.h>
130 
131 #ifdef COMPAT_IA32
132 #include <sys/mount.h>
133 #include <compat/freebsd32/freebsd32.h>
134 
135 extern struct sysentvec ia32_freebsd_sysvec;
136 #endif
137 
138 static int	soreceive_rcvoob(struct socket *so, struct uio *uio,
139 		    int flags);
140 
141 static void	filt_sordetach(struct knote *kn);
142 static int	filt_soread(struct knote *kn, long hint);
143 static void	filt_sowdetach(struct knote *kn);
144 static int	filt_sowrite(struct knote *kn, long hint);
145 static int	filt_solisten(struct knote *kn, long hint);
146 
147 static struct filterops solisten_filtops =
148 	{ 1, NULL, filt_sordetach, filt_solisten };
149 static struct filterops soread_filtops =
150 	{ 1, NULL, filt_sordetach, filt_soread };
151 static struct filterops sowrite_filtops =
152 	{ 1, NULL, filt_sowdetach, filt_sowrite };
153 
154 uma_zone_t socket_zone;
155 so_gen_t	so_gencnt;	/* generation count for sockets */
156 
157 int	maxsockets;
158 
159 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
160 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
161 
162 static int somaxconn = SOMAXCONN;
163 static int somaxconn_sysctl(SYSCTL_HANDLER_ARGS);
164 /* XXX: we dont have SYSCTL_USHORT */
165 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLTYPE_UINT | CTLFLAG_RW,
166     0, sizeof(int), somaxconn_sysctl, "I", "Maximum pending socket connection "
167     "queue size");
168 static int numopensockets;
169 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
170     &numopensockets, 0, "Number of open sockets");
171 #ifdef ZERO_COPY_SOCKETS
172 /* These aren't static because they're used in other files. */
173 int so_zero_copy_send = 1;
174 int so_zero_copy_receive = 1;
175 SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0,
176     "Zero copy controls");
177 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW,
178     &so_zero_copy_receive, 0, "Enable zero copy receive");
179 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW,
180     &so_zero_copy_send, 0, "Enable zero copy send");
181 #endif /* ZERO_COPY_SOCKETS */
182 
183 /*
184  * accept_mtx locks down per-socket fields relating to accept queues.  See
185  * socketvar.h for an annotation of the protected fields of struct socket.
186  */
187 struct mtx accept_mtx;
188 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
189 
190 /*
191  * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
192  * so_gencnt field.
193  */
194 static struct mtx so_global_mtx;
195 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
196 
197 /*
198  * General IPC sysctl name space, used by sockets and a variety of other IPC
199  * types.
200  */
201 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
202 
203 /*
204  * Sysctl to get and set the maximum global sockets limit.  Notify protocols
205  * of the change so that they can update their dependent limits as required.
206  */
207 static int
208 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
209 {
210 	int error, newmaxsockets;
211 
212 	newmaxsockets = maxsockets;
213 	error = sysctl_handle_int(oidp, &newmaxsockets, sizeof(int), req);
214 	if (error == 0 && req->newptr) {
215 		if (newmaxsockets > maxsockets) {
216 			maxsockets = newmaxsockets;
217 			if (maxsockets > ((maxfiles / 4) * 3)) {
218 				maxfiles = (maxsockets * 5) / 4;
219 				maxfilesperproc = (maxfiles * 9) / 10;
220 			}
221 			EVENTHANDLER_INVOKE(maxsockets_change);
222 		} else
223 			error = EINVAL;
224 	}
225 	return (error);
226 }
227 
228 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
229     &maxsockets, 0, sysctl_maxsockets, "IU",
230     "Maximum number of sockets avaliable");
231 
232 /*
233  * Initialise maxsockets.
234  */
235 static void init_maxsockets(void *ignored)
236 {
237 	TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
238 	maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
239 }
240 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
241 
242 /*
243  * Socket operation routines.  These routines are called by the routines in
244  * sys_socket.c or from a system process, and implement the semantics of
245  * socket operations by switching out to the protocol specific routines.
246  */
247 
248 /*
249  * Get a socket structure from our zone, and initialize it.  Note that it
250  * would probably be better to allocate socket and PCB at the same time, but
251  * I'm not convinced that all the protocols can be easily modified to do
252  * this.
253  *
254  * soalloc() returns a socket with a ref count of 0.
255  */
256 static struct socket *
257 soalloc(int mflags)
258 {
259 	struct socket *so;
260 
261 	so = uma_zalloc(socket_zone, mflags | M_ZERO);
262 	if (so == NULL)
263 		return (NULL);
264 #ifdef MAC
265 	if (mac_init_socket(so, mflags) != 0) {
266 		uma_zfree(socket_zone, so);
267 		return (NULL);
268 	}
269 #endif
270 	SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
271 	SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
272 	TAILQ_INIT(&so->so_aiojobq);
273 	mtx_lock(&so_global_mtx);
274 	so->so_gencnt = ++so_gencnt;
275 	++numopensockets;
276 	mtx_unlock(&so_global_mtx);
277 	return (so);
278 }
279 
280 /*
281  * Free the storage associated with a socket at the socket layer, tear down
282  * locks, labels, etc.  All protocol state is assumed already to have been
283  * torn down (and possibly never set up) by the caller.
284  */
285 static void
286 sodealloc(struct socket *so)
287 {
288 
289 	KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
290 	KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
291 
292 	mtx_lock(&so_global_mtx);
293 	so->so_gencnt = ++so_gencnt;
294 	--numopensockets;	/* Could be below, but faster here. */
295 	mtx_unlock(&so_global_mtx);
296 	if (so->so_rcv.sb_hiwat)
297 		(void)chgsbsize(so->so_cred->cr_uidinfo,
298 		    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
299 	if (so->so_snd.sb_hiwat)
300 		(void)chgsbsize(so->so_cred->cr_uidinfo,
301 		    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
302 #ifdef INET
303 	/* remove acccept filter if one is present. */
304 	if (so->so_accf != NULL)
305 		do_setopt_accept_filter(so, NULL);
306 #endif
307 #ifdef MAC
308 	mac_destroy_socket(so);
309 #endif
310 	crfree(so->so_cred);
311 	SOCKBUF_LOCK_DESTROY(&so->so_snd);
312 	SOCKBUF_LOCK_DESTROY(&so->so_rcv);
313 	uma_zfree(socket_zone, so);
314 }
315 
316 /*
317  * socreate returns a socket with a ref count of 1.  The socket should be
318  * closed with soclose().
319  */
320 int
321 socreate(dom, aso, type, proto, cred, td)
322 	int dom;
323 	struct socket **aso;
324 	int type;
325 	int proto;
326 	struct ucred *cred;
327 	struct thread *td;
328 {
329 	struct protosw *prp;
330 	struct socket *so;
331 	int error;
332 
333 	if (proto)
334 		prp = pffindproto(dom, proto, type);
335 	else
336 		prp = pffindtype(dom, type);
337 
338 	if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL ||
339 	    prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
340 		return (EPROTONOSUPPORT);
341 
342 	if (jailed(cred) && jail_socket_unixiproute_only &&
343 	    prp->pr_domain->dom_family != PF_LOCAL &&
344 	    prp->pr_domain->dom_family != PF_INET &&
345 	    prp->pr_domain->dom_family != PF_ROUTE) {
346 		return (EPROTONOSUPPORT);
347 	}
348 
349 	if (prp->pr_type != type)
350 		return (EPROTOTYPE);
351 	so = soalloc(M_WAITOK);
352 	if (so == NULL)
353 		return (ENOBUFS);
354 
355 	TAILQ_INIT(&so->so_incomp);
356 	TAILQ_INIT(&so->so_comp);
357 	so->so_type = type;
358 	so->so_cred = crhold(cred);
359 	so->so_proto = prp;
360 #ifdef MAC
361 	mac_create_socket(cred, so);
362 #endif
363 	knlist_init(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv),
364 	    NULL, NULL, NULL);
365 	knlist_init(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd),
366 	    NULL, NULL, NULL);
367 	so->so_count = 1;
368 	error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
369 	if (error) {
370 		KASSERT(so->so_count == 1, ("socreate: so_count %d",
371 		    so->so_count));
372 		so->so_count = 0;
373 		sodealloc(so);
374 		return (error);
375 	}
376 	*aso = so;
377 	return (0);
378 }
379 
380 #ifdef REGRESSION
381 static int regression_sonewconn_earlytest = 1;
382 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
383     &regression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
384 #endif
385 
386 /*
387  * When an attempt at a new connection is noted on a socket which accepts
388  * connections, sonewconn is called.  If the connection is possible (subject
389  * to space constraints, etc.) then we allocate a new structure, propoerly
390  * linked into the data structure of the original socket, and return this.
391  * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
392  *
393  * Note: the ref count on the socket is 0 on return.
394  */
395 struct socket *
396 sonewconn(head, connstatus)
397 	register struct socket *head;
398 	int connstatus;
399 {
400 	register struct socket *so;
401 	int over;
402 
403 	ACCEPT_LOCK();
404 	over = (head->so_qlen > 3 * head->so_qlimit / 2);
405 	ACCEPT_UNLOCK();
406 #ifdef REGRESSION
407 	if (regression_sonewconn_earlytest && over)
408 #else
409 	if (over)
410 #endif
411 		return (NULL);
412 	so = soalloc(M_NOWAIT);
413 	if (so == NULL)
414 		return (NULL);
415 	if ((head->so_options & SO_ACCEPTFILTER) != 0)
416 		connstatus = 0;
417 	so->so_head = head;
418 	so->so_type = head->so_type;
419 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
420 	so->so_linger = head->so_linger;
421 	so->so_state = head->so_state | SS_NOFDREF;
422 	so->so_proto = head->so_proto;
423 	so->so_timeo = head->so_timeo;
424 	so->so_cred = crhold(head->so_cred);
425 #ifdef MAC
426 	SOCK_LOCK(head);
427 	mac_create_socket_from_socket(head, so);
428 	SOCK_UNLOCK(head);
429 #endif
430 	knlist_init(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv),
431 	    NULL, NULL, NULL);
432 	knlist_init(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd),
433 	    NULL, NULL, NULL);
434 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
435 	    (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
436 		sodealloc(so);
437 		return (NULL);
438 	}
439 	so->so_state |= connstatus;
440 	ACCEPT_LOCK();
441 	if (connstatus) {
442 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
443 		so->so_qstate |= SQ_COMP;
444 		head->so_qlen++;
445 	} else {
446 		/*
447 		 * Keep removing sockets from the head until there's room for
448 		 * us to insert on the tail.  In pre-locking revisions, this
449 		 * was a simple if(), but as we could be racing with other
450 		 * threads and soabort() requires dropping locks, we must
451 		 * loop waiting for the condition to be true.
452 		 */
453 		while (head->so_incqlen > head->so_qlimit) {
454 			struct socket *sp;
455 			sp = TAILQ_FIRST(&head->so_incomp);
456 			TAILQ_REMOVE(&head->so_incomp, sp, so_list);
457 			head->so_incqlen--;
458 			sp->so_qstate &= ~SQ_INCOMP;
459 			sp->so_head = NULL;
460 			ACCEPT_UNLOCK();
461 			soabort(sp);
462 			ACCEPT_LOCK();
463 		}
464 		TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
465 		so->so_qstate |= SQ_INCOMP;
466 		head->so_incqlen++;
467 	}
468 	ACCEPT_UNLOCK();
469 	if (connstatus) {
470 		sorwakeup(head);
471 		wakeup_one(&head->so_timeo);
472 	}
473 	return (so);
474 }
475 
476 int
477 sobind(so, nam, td)
478 	struct socket *so;
479 	struct sockaddr *nam;
480 	struct thread *td;
481 {
482 
483 	return ((*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td));
484 }
485 
486 /*
487  * solisten() transitions a socket from a non-listening state to a listening
488  * state, but can also be used to update the listen queue depth on an
489  * existing listen socket.  The protocol will call back into the sockets
490  * layer using solisten_proto_check() and solisten_proto() to check and set
491  * socket-layer listen state.  Call backs are used so that the protocol can
492  * acquire both protocol and socket layer locks in whatever order is required
493  * by the protocol.
494  *
495  * Protocol implementors are advised to hold the socket lock across the
496  * socket-layer test and set to avoid races at the socket layer.
497  */
498 int
499 solisten(so, backlog, td)
500 	struct socket *so;
501 	int backlog;
502 	struct thread *td;
503 {
504 
505 	return ((*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td));
506 }
507 
508 int
509 solisten_proto_check(so)
510 	struct socket *so;
511 {
512 
513 	SOCK_LOCK_ASSERT(so);
514 
515 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
516 	    SS_ISDISCONNECTING))
517 		return (EINVAL);
518 	return (0);
519 }
520 
521 void
522 solisten_proto(so, backlog)
523 	struct socket *so;
524 	int backlog;
525 {
526 
527 	SOCK_LOCK_ASSERT(so);
528 
529 	if (backlog < 0 || backlog > somaxconn)
530 		backlog = somaxconn;
531 	so->so_qlimit = backlog;
532 	so->so_options |= SO_ACCEPTCONN;
533 }
534 
535 /*
536  * Attempt to free a socket.  This should really be sotryfree().
537  *
538  * sofree() will succeed if:
539  *
540  * - There are no outstanding file descriptor references or related consumers
541  *   (so_count == 0).
542  *
543  * - The socket has been closed by user space, if ever open (SS_NOFDREF).
544  *
545  * - The protocol does not have an outstanding strong reference on the socket
546  *   (SS_PROTOREF).
547  *
548  * - The socket is not in a completed connection queue, so a process has been
549  *   notified that it is present.  If it is removed, the user process may
550  *   block in accept() despite select() saying the socket was ready.
551  *
552  * Otherwise, it will quietly abort so that a future call to sofree(), when
553  * conditions are right, can succeed.
554  */
555 void
556 sofree(so)
557 	struct socket *so;
558 {
559 	struct protosw *pr = so->so_proto;
560 	struct socket *head;
561 
562 	ACCEPT_LOCK_ASSERT();
563 	SOCK_LOCK_ASSERT(so);
564 
565 	if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
566 	    (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
567 		SOCK_UNLOCK(so);
568 		ACCEPT_UNLOCK();
569 		return;
570 	}
571 
572 	head = so->so_head;
573 	if (head != NULL) {
574 		KASSERT((so->so_qstate & SQ_COMP) != 0 ||
575 		    (so->so_qstate & SQ_INCOMP) != 0,
576 		    ("sofree: so_head != NULL, but neither SQ_COMP nor "
577 		    "SQ_INCOMP"));
578 		KASSERT((so->so_qstate & SQ_COMP) == 0 ||
579 		    (so->so_qstate & SQ_INCOMP) == 0,
580 		    ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
581 		TAILQ_REMOVE(&head->so_incomp, so, so_list);
582 		head->so_incqlen--;
583 		so->so_qstate &= ~SQ_INCOMP;
584 		so->so_head = NULL;
585 	}
586 	KASSERT((so->so_qstate & SQ_COMP) == 0 &&
587 	    (so->so_qstate & SQ_INCOMP) == 0,
588 	    ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
589 	    so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
590 	SOCK_UNLOCK(so);
591 	ACCEPT_UNLOCK();
592 
593 	/*
594 	 * From this point on, we assume that no other references to this
595 	 * socket exist anywhere else in the stack.  Therefore, no locks need
596 	 * to be acquired or held.
597 	 *
598 	 * We used to do a lot of socket buffer and socket locking here, as
599 	 * well as invoke sorflush() and perform wakeups.  The direct call to
600 	 * dom_dispose() and sbrelease_internal() are an inlining of what was
601 	 * necessary from sorflush().
602 	 *
603 	 * Notice that the socket buffer and kqueue state are torn down
604 	 * before calling pru_detach.  This means that protocols shold not
605 	 * assume they can perform socket wakeups, etc, in their detach
606 	 * code.
607 	 */
608 	KASSERT((so->so_snd.sb_flags & SB_LOCK) == 0, ("sofree: snd sblock"));
609 	KASSERT((so->so_rcv.sb_flags & SB_LOCK) == 0, ("sofree: rcv sblock"));
610 	sbdestroy(&so->so_snd, so);
611 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
612 		(*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
613 	sbdestroy(&so->so_rcv, so);
614 	if (pr->pr_usrreqs->pru_detach != NULL)
615 		(*pr->pr_usrreqs->pru_detach)(so);
616 	knlist_destroy(&so->so_rcv.sb_sel.si_note);
617 	knlist_destroy(&so->so_snd.sb_sel.si_note);
618 	sodealloc(so);
619 }
620 
621 /*
622  * Close a socket on last file table reference removal.  Initiate disconnect
623  * if connected.  Free socket when disconnect complete.
624  *
625  * This function will sorele() the socket.  Note that soclose() may be called
626  * prior to the ref count reaching zero.  The actual socket structure will
627  * not be freed until the ref count reaches zero.
628  */
629 int
630 soclose(so)
631 	struct socket *so;
632 {
633 	int error = 0;
634 
635 	KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
636 
637 	funsetown(&so->so_sigio);
638 	if (so->so_options & SO_ACCEPTCONN) {
639 		struct socket *sp;
640 		ACCEPT_LOCK();
641 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
642 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
643 			so->so_incqlen--;
644 			sp->so_qstate &= ~SQ_INCOMP;
645 			sp->so_head = NULL;
646 			ACCEPT_UNLOCK();
647 			soabort(sp);
648 			ACCEPT_LOCK();
649 		}
650 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
651 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
652 			so->so_qlen--;
653 			sp->so_qstate &= ~SQ_COMP;
654 			sp->so_head = NULL;
655 			ACCEPT_UNLOCK();
656 			soabort(sp);
657 			ACCEPT_LOCK();
658 		}
659 		ACCEPT_UNLOCK();
660 	}
661 	if (so->so_state & SS_ISCONNECTED) {
662 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
663 			error = sodisconnect(so);
664 			if (error)
665 				goto drop;
666 		}
667 		if (so->so_options & SO_LINGER) {
668 			if ((so->so_state & SS_ISDISCONNECTING) &&
669 			    (so->so_state & SS_NBIO))
670 				goto drop;
671 			while (so->so_state & SS_ISCONNECTED) {
672 				error = tsleep(&so->so_timeo,
673 				    PSOCK | PCATCH, "soclos", so->so_linger * hz);
674 				if (error)
675 					break;
676 			}
677 		}
678 	}
679 
680 drop:
681 	if (so->so_proto->pr_usrreqs->pru_close != NULL)
682 		(*so->so_proto->pr_usrreqs->pru_close)(so);
683 	ACCEPT_LOCK();
684 	SOCK_LOCK(so);
685 	KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
686 	so->so_state |= SS_NOFDREF;
687 	sorele(so);
688 	return (error);
689 }
690 
691 /*
692  * soabort() is used to abruptly tear down a connection, such as when a
693  * resource limit is reached (listen queue depth exceeded), or if a listen
694  * socket is closed while there are sockets waiting to be accepted.
695  *
696  * This interface is tricky, because it is called on an unreferenced socket,
697  * and must be called only by a thread that has actually removed the socket
698  * from the listen queue it was on, or races with other threads are risked.
699  *
700  * This interface will call into the protocol code, so must not be called
701  * with any socket locks held.  Protocols do call it while holding their own
702  * recursible protocol mutexes, but this is something that should be subject
703  * to review in the future.
704  */
705 void
706 soabort(so)
707 	struct socket *so;
708 {
709 
710 	/*
711 	 * In as much as is possible, assert that no references to this
712 	 * socket are held.  This is not quite the same as asserting that the
713 	 * current thread is responsible for arranging for no references, but
714 	 * is as close as we can get for now.
715 	 */
716 	KASSERT(so->so_count == 0, ("soabort: so_count"));
717 	KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
718 	KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
719 	KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
720 	KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
721 
722 	if (so->so_proto->pr_usrreqs->pru_abort != NULL)
723 		(*so->so_proto->pr_usrreqs->pru_abort)(so);
724 	ACCEPT_LOCK();
725 	SOCK_LOCK(so);
726 	sofree(so);
727 }
728 
729 int
730 soaccept(so, nam)
731 	struct socket *so;
732 	struct sockaddr **nam;
733 {
734 	int error;
735 
736 	SOCK_LOCK(so);
737 	KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
738 	so->so_state &= ~SS_NOFDREF;
739 	SOCK_UNLOCK(so);
740 	error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
741 	return (error);
742 }
743 
744 int
745 soconnect(so, nam, td)
746 	struct socket *so;
747 	struct sockaddr *nam;
748 	struct thread *td;
749 {
750 	int error;
751 
752 	if (so->so_options & SO_ACCEPTCONN)
753 		return (EOPNOTSUPP);
754 	/*
755 	 * If protocol is connection-based, can only connect once.
756 	 * Otherwise, if connected, try to disconnect first.  This allows
757 	 * user to disconnect by connecting to, e.g., a null address.
758 	 */
759 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
760 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
761 	    (error = sodisconnect(so)))) {
762 		error = EISCONN;
763 	} else {
764 		/*
765 		 * Prevent accumulated error from previous connection from
766 		 * biting us.
767 		 */
768 		so->so_error = 0;
769 		error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td);
770 	}
771 
772 	return (error);
773 }
774 
775 int
776 soconnect2(so1, so2)
777 	struct socket *so1;
778 	struct socket *so2;
779 {
780 
781 	return ((*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2));
782 }
783 
784 int
785 sodisconnect(so)
786 	struct socket *so;
787 {
788 	int error;
789 
790 	if ((so->so_state & SS_ISCONNECTED) == 0)
791 		return (ENOTCONN);
792 	if (so->so_state & SS_ISDISCONNECTING)
793 		return (EALREADY);
794 	error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
795 	return (error);
796 }
797 
798 #ifdef ZERO_COPY_SOCKETS
799 struct so_zerocopy_stats{
800 	int size_ok;
801 	int align_ok;
802 	int found_ifp;
803 };
804 struct so_zerocopy_stats so_zerocp_stats = {0,0,0};
805 #include <netinet/in.h>
806 #include <net/route.h>
807 #include <netinet/in_pcb.h>
808 #include <vm/vm.h>
809 #include <vm/vm_page.h>
810 #include <vm/vm_object.h>
811 #endif /*ZERO_COPY_SOCKETS*/
812 
813 /*
814  * sosend_copyin() accepts a uio and prepares an mbuf chain holding part or
815  * all of the data referenced by the uio.  If desired, it uses zero-copy.
816  * *space will be updated to reflect data copied in.
817  *
818  * NB: If atomic I/O is requested, the caller must already have checked that
819  * space can hold resid bytes.
820  *
821  * NB: In the event of an error, the caller may need to free the partial
822  * chain pointed to by *mpp.  The contents of both *uio and *space may be
823  * modified even in the case of an error.
824  */
825 static int
826 sosend_copyin(struct uio *uio, struct mbuf **retmp, int atomic, long *space,
827     int flags)
828 {
829 	struct mbuf *m, **mp, *top;
830 	long len, resid;
831 	int error;
832 #ifdef ZERO_COPY_SOCKETS
833 	int cow_send;
834 #endif
835 
836 	*retmp = top = NULL;
837 	mp = &top;
838 	len = 0;
839 	resid = uio->uio_resid;
840 	error = 0;
841 	do {
842 #ifdef ZERO_COPY_SOCKETS
843 		cow_send = 0;
844 #endif /* ZERO_COPY_SOCKETS */
845 		if (resid >= MINCLSIZE) {
846 #ifdef ZERO_COPY_SOCKETS
847 			if (top == NULL) {
848 				MGETHDR(m, M_TRYWAIT, MT_DATA);
849 				if (m == NULL) {
850 					error = ENOBUFS;
851 					goto out;
852 				}
853 				m->m_pkthdr.len = 0;
854 				m->m_pkthdr.rcvif = NULL;
855 			} else {
856 				MGET(m, M_TRYWAIT, MT_DATA);
857 				if (m == NULL) {
858 					error = ENOBUFS;
859 					goto out;
860 				}
861 			}
862 			if (so_zero_copy_send &&
863 			    resid>=PAGE_SIZE &&
864 			    *space>=PAGE_SIZE &&
865 			    uio->uio_iov->iov_len>=PAGE_SIZE) {
866 				so_zerocp_stats.size_ok++;
867 				so_zerocp_stats.align_ok++;
868 				cow_send = socow_setup(m, uio);
869 				len = cow_send;
870 			}
871 			if (!cow_send) {
872 				MCLGET(m, M_TRYWAIT);
873 				if ((m->m_flags & M_EXT) == 0) {
874 					m_free(m);
875 					m = NULL;
876 				} else {
877 					len = min(min(MCLBYTES, resid),
878 					    *space);
879 				}
880 			}
881 #else /* ZERO_COPY_SOCKETS */
882 			if (top == NULL) {
883 				m = m_getcl(M_TRYWAIT, MT_DATA, M_PKTHDR);
884 				m->m_pkthdr.len = 0;
885 				m->m_pkthdr.rcvif = NULL;
886 			} else
887 				m = m_getcl(M_TRYWAIT, MT_DATA, 0);
888 			len = min(min(MCLBYTES, resid), *space);
889 #endif /* ZERO_COPY_SOCKETS */
890 		} else {
891 			if (top == NULL) {
892 				m = m_gethdr(M_TRYWAIT, MT_DATA);
893 				m->m_pkthdr.len = 0;
894 				m->m_pkthdr.rcvif = NULL;
895 
896 				len = min(min(MHLEN, resid), *space);
897 				/*
898 				 * For datagram protocols, leave room
899 				 * for protocol headers in first mbuf.
900 				 */
901 				if (atomic && m && len < MHLEN)
902 					MH_ALIGN(m, len);
903 			} else {
904 				m = m_get(M_TRYWAIT, MT_DATA);
905 				len = min(min(MLEN, resid), *space);
906 			}
907 		}
908 		if (m == NULL) {
909 			error = ENOBUFS;
910 			goto out;
911 		}
912 
913 		*space -= len;
914 #ifdef ZERO_COPY_SOCKETS
915 		if (cow_send)
916 			error = 0;
917 		else
918 #endif /* ZERO_COPY_SOCKETS */
919 		error = uiomove(mtod(m, void *), (int)len, uio);
920 		resid = uio->uio_resid;
921 		m->m_len = len;
922 		*mp = m;
923 		top->m_pkthdr.len += len;
924 		if (error)
925 			goto out;
926 		mp = &m->m_next;
927 		if (resid <= 0) {
928 			if (flags & MSG_EOR)
929 				top->m_flags |= M_EOR;
930 			break;
931 		}
932 	} while (*space > 0 && atomic);
933 out:
934 	*retmp = top;
935 	return (error);
936 }
937 
938 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
939 
940 int
941 sosend_dgram(so, addr, uio, top, control, flags, td)
942 	struct socket *so;
943 	struct sockaddr *addr;
944 	struct uio *uio;
945 	struct mbuf *top;
946 	struct mbuf *control;
947 	int flags;
948 	struct thread *td;
949 {
950 	long space, resid;
951 	int clen = 0, error, dontroute;
952 	int atomic = sosendallatonce(so) || top;
953 
954 	KASSERT(so->so_type == SOCK_DGRAM, ("sodgram_send: !SOCK_DGRAM"));
955 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
956 	    ("sodgram_send: !PR_ATOMIC"));
957 
958 	if (uio != NULL)
959 		resid = uio->uio_resid;
960 	else
961 		resid = top->m_pkthdr.len;
962 	/*
963 	 * In theory resid should be unsigned.  However, space must be
964 	 * signed, as it might be less than 0 if we over-committed, and we
965 	 * must use a signed comparison of space and resid.  On the other
966 	 * hand, a negative resid causes us to loop sending 0-length
967 	 * segments to the protocol.
968 	 *
969 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
970 	 * type sockets since that's an error.
971 	 */
972 	if (resid < 0) {
973 		error = EINVAL;
974 		goto out;
975 	}
976 
977 	dontroute =
978 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
979 	if (td != NULL)
980 		td->td_proc->p_stats->p_ru.ru_msgsnd++;
981 	if (control != NULL)
982 		clen = control->m_len;
983 
984 	SOCKBUF_LOCK(&so->so_snd);
985 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
986 		SOCKBUF_UNLOCK(&so->so_snd);
987 		error = EPIPE;
988 		goto out;
989 	}
990 	if (so->so_error) {
991 		error = so->so_error;
992 		so->so_error = 0;
993 		SOCKBUF_UNLOCK(&so->so_snd);
994 		goto out;
995 	}
996 	if ((so->so_state & SS_ISCONNECTED) == 0) {
997 		/*
998 		 * `sendto' and `sendmsg' is allowed on a connection-based
999 		 * socket if it supports implied connect.  Return ENOTCONN if
1000 		 * not connected and no address is supplied.
1001 		 */
1002 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1003 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1004 			if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1005 			    !(resid == 0 && clen != 0)) {
1006 				SOCKBUF_UNLOCK(&so->so_snd);
1007 				error = ENOTCONN;
1008 				goto out;
1009 			}
1010 		} else if (addr == NULL) {
1011 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1012 				error = ENOTCONN;
1013 			else
1014 				error = EDESTADDRREQ;
1015 			SOCKBUF_UNLOCK(&so->so_snd);
1016 			goto out;
1017 		}
1018 	}
1019 
1020 	/*
1021 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
1022 	 * problem and need fixing.
1023 	 */
1024 	space = sbspace(&so->so_snd);
1025 	if (flags & MSG_OOB)
1026 		space += 1024;
1027 	space -= clen;
1028 	if (resid > space) {
1029 		error = EMSGSIZE;
1030 		goto out;
1031 	}
1032 	SOCKBUF_UNLOCK(&so->so_snd);
1033 	if (uio == NULL) {
1034 		resid = 0;
1035 		if (flags & MSG_EOR)
1036 			top->m_flags |= M_EOR;
1037 	} else {
1038 		error = sosend_copyin(uio, &top, atomic, &space, flags);
1039 		if (error)
1040 			goto out;
1041 		resid = uio->uio_resid;
1042 	}
1043 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1044 	/*
1045 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1046 	 * than with.
1047 	 */
1048 	if (dontroute) {
1049 		SOCK_LOCK(so);
1050 		so->so_options |= SO_DONTROUTE;
1051 		SOCK_UNLOCK(so);
1052 	}
1053 	/*
1054 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1055 	 * of date.  We could have recieved a reset packet in an interrupt or
1056 	 * maybe we slept while doing page faults in uiomove() etc.  We could
1057 	 * probably recheck again inside the locking protection here, but
1058 	 * there are probably other places that this also happens.  We must
1059 	 * rethink this.
1060 	 */
1061 	error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1062 	    (flags & MSG_OOB) ? PRUS_OOB :
1063 	/*
1064 	 * If the user set MSG_EOF, the protocol understands this flag and
1065 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1066 	 */
1067 	    ((flags & MSG_EOF) &&
1068 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1069 	     (resid <= 0)) ?
1070 		PRUS_EOF :
1071 		/* If there is more to send set PRUS_MORETOCOME */
1072 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1073 		top, addr, control, td);
1074 	if (dontroute) {
1075 		SOCK_LOCK(so);
1076 		so->so_options &= ~SO_DONTROUTE;
1077 		SOCK_UNLOCK(so);
1078 	}
1079 	clen = 0;
1080 	control = NULL;
1081 	top = NULL;
1082 out:
1083 	if (top != NULL)
1084 		m_freem(top);
1085 	if (control != NULL)
1086 		m_freem(control);
1087 	return (error);
1088 }
1089 
1090 /*
1091  * Send on a socket.  If send must go all at once and message is larger than
1092  * send buffering, then hard error.  Lock against other senders.  If must go
1093  * all at once and not enough room now, then inform user that this would
1094  * block and do nothing.  Otherwise, if nonblocking, send as much as
1095  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1096  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1097  * in mbuf chain must be small enough to send all at once.
1098  *
1099  * Returns nonzero on error, timeout or signal; callers must check for short
1100  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1101  * on return.
1102  */
1103 #define	snderr(errno)	{ error = (errno); goto release; }
1104 int
1105 sosend_generic(so, addr, uio, top, control, flags, td)
1106 	struct socket *so;
1107 	struct sockaddr *addr;
1108 	struct uio *uio;
1109 	struct mbuf *top;
1110 	struct mbuf *control;
1111 	int flags;
1112 	struct thread *td;
1113 {
1114 	long space, resid;
1115 	int clen = 0, error, dontroute;
1116 	int atomic = sosendallatonce(so) || top;
1117 
1118 	if (uio != NULL)
1119 		resid = uio->uio_resid;
1120 	else
1121 		resid = top->m_pkthdr.len;
1122 	/*
1123 	 * In theory resid should be unsigned.  However, space must be
1124 	 * signed, as it might be less than 0 if we over-committed, and we
1125 	 * must use a signed comparison of space and resid.  On the other
1126 	 * hand, a negative resid causes us to loop sending 0-length
1127 	 * segments to the protocol.
1128 	 *
1129 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1130 	 * type sockets since that's an error.
1131 	 */
1132 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1133 		error = EINVAL;
1134 		goto out;
1135 	}
1136 
1137 	dontroute =
1138 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1139 	    (so->so_proto->pr_flags & PR_ATOMIC);
1140 	if (td != NULL)
1141 		td->td_proc->p_stats->p_ru.ru_msgsnd++;
1142 	if (control != NULL)
1143 		clen = control->m_len;
1144 
1145 	SOCKBUF_LOCK(&so->so_snd);
1146 restart:
1147 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
1148 	error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1149 	if (error)
1150 		goto out_locked;
1151 	do {
1152 		SOCKBUF_LOCK_ASSERT(&so->so_snd);
1153 		if (so->so_snd.sb_state & SBS_CANTSENDMORE)
1154 			snderr(EPIPE);
1155 		if (so->so_error) {
1156 			error = so->so_error;
1157 			so->so_error = 0;
1158 			goto release;
1159 		}
1160 		if ((so->so_state & SS_ISCONNECTED) == 0) {
1161 			/*
1162 			 * `sendto' and `sendmsg' is allowed on a connection-
1163 			 * based socket if it supports implied connect.
1164 			 * Return ENOTCONN if not connected and no address is
1165 			 * supplied.
1166 			 */
1167 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1168 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1169 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1170 				    !(resid == 0 && clen != 0))
1171 					snderr(ENOTCONN);
1172 			} else if (addr == NULL)
1173 			    snderr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
1174 				   ENOTCONN : EDESTADDRREQ);
1175 		}
1176 		space = sbspace(&so->so_snd);
1177 		if (flags & MSG_OOB)
1178 			space += 1024;
1179 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1180 		    clen > so->so_snd.sb_hiwat)
1181 			snderr(EMSGSIZE);
1182 		if (space < resid + clen &&
1183 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1184 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO))
1185 				snderr(EWOULDBLOCK);
1186 			sbunlock(&so->so_snd);
1187 			error = sbwait(&so->so_snd);
1188 			if (error)
1189 				goto out_locked;
1190 			goto restart;
1191 		}
1192 		SOCKBUF_UNLOCK(&so->so_snd);
1193 		space -= clen;
1194 		do {
1195 			if (uio == NULL) {
1196 				resid = 0;
1197 				if (flags & MSG_EOR)
1198 					top->m_flags |= M_EOR;
1199 			} else {
1200 				error = sosend_copyin(uio, &top, atomic,
1201 				    &space, flags);
1202 				if (error != 0) {
1203 					SOCKBUF_LOCK(&so->so_snd);
1204 					goto release;
1205 				}
1206 				resid = uio->uio_resid;
1207 			}
1208 			if (dontroute) {
1209 				SOCK_LOCK(so);
1210 				so->so_options |= SO_DONTROUTE;
1211 				SOCK_UNLOCK(so);
1212 			}
1213 			/*
1214 			 * XXX all the SBS_CANTSENDMORE checks previously
1215 			 * done could be out of date.  We could have recieved
1216 			 * a reset packet in an interrupt or maybe we slept
1217 			 * while doing page faults in uiomove() etc.  We
1218 			 * could probably recheck again inside the locking
1219 			 * protection here, but there are probably other
1220 			 * places that this also happens.  We must rethink
1221 			 * this.
1222 			 */
1223 			error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1224 			    (flags & MSG_OOB) ? PRUS_OOB :
1225 			/*
1226 			 * If the user set MSG_EOF, the protocol understands
1227 			 * this flag and nothing left to send then use
1228 			 * PRU_SEND_EOF instead of PRU_SEND.
1229 			 */
1230 			    ((flags & MSG_EOF) &&
1231 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1232 			     (resid <= 0)) ?
1233 				PRUS_EOF :
1234 			/* If there is more to send set PRUS_MORETOCOME. */
1235 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1236 			    top, addr, control, td);
1237 			if (dontroute) {
1238 				SOCK_LOCK(so);
1239 				so->so_options &= ~SO_DONTROUTE;
1240 				SOCK_UNLOCK(so);
1241 			}
1242 			clen = 0;
1243 			control = NULL;
1244 			top = NULL;
1245 			if (error) {
1246 				SOCKBUF_LOCK(&so->so_snd);
1247 				goto release;
1248 			}
1249 		} while (resid && space > 0);
1250 		SOCKBUF_LOCK(&so->so_snd);
1251 	} while (resid);
1252 
1253 release:
1254 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
1255 	sbunlock(&so->so_snd);
1256 out_locked:
1257 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
1258 	SOCKBUF_UNLOCK(&so->so_snd);
1259 out:
1260 	if (top != NULL)
1261 		m_freem(top);
1262 	if (control != NULL)
1263 		m_freem(control);
1264 	return (error);
1265 }
1266 #undef snderr
1267 
1268 int
1269 sosend(so, addr, uio, top, control, flags, td)
1270 	struct socket *so;
1271 	struct sockaddr *addr;
1272 	struct uio *uio;
1273 	struct mbuf *top;
1274 	struct mbuf *control;
1275 	int flags;
1276 	struct thread *td;
1277 {
1278 
1279 	/* XXXRW: Temporary debugging. */
1280 	KASSERT(so->so_proto->pr_usrreqs->pru_sosend != sosend,
1281 	    ("sosend: protocol calls sosend"));
1282 
1283 	return (so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1284 	    control, flags, td));
1285 }
1286 
1287 /*
1288  * The part of soreceive() that implements reading non-inline out-of-band
1289  * data from a socket.  For more complete comments, see soreceive(), from
1290  * which this code originated.
1291  *
1292  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1293  * unable to return an mbuf chain to the caller.
1294  */
1295 static int
1296 soreceive_rcvoob(so, uio, flags)
1297 	struct socket *so;
1298 	struct uio *uio;
1299 	int flags;
1300 {
1301 	struct protosw *pr = so->so_proto;
1302 	struct mbuf *m;
1303 	int error;
1304 
1305 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1306 
1307 	m = m_get(M_TRYWAIT, MT_DATA);
1308 	if (m == NULL)
1309 		return (ENOBUFS);
1310 	error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1311 	if (error)
1312 		goto bad;
1313 	do {
1314 #ifdef ZERO_COPY_SOCKETS
1315 		if (so_zero_copy_receive) {
1316 			int disposable;
1317 
1318 			if ((m->m_flags & M_EXT)
1319 			 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1320 				disposable = 1;
1321 			else
1322 				disposable = 0;
1323 
1324 			error = uiomoveco(mtod(m, void *),
1325 					  min(uio->uio_resid, m->m_len),
1326 					  uio, disposable);
1327 		} else
1328 #endif /* ZERO_COPY_SOCKETS */
1329 		error = uiomove(mtod(m, void *),
1330 		    (int) min(uio->uio_resid, m->m_len), uio);
1331 		m = m_free(m);
1332 	} while (uio->uio_resid && error == 0 && m);
1333 bad:
1334 	if (m != NULL)
1335 		m_freem(m);
1336 	return (error);
1337 }
1338 
1339 /*
1340  * Following replacement or removal of the first mbuf on the first mbuf chain
1341  * of a socket buffer, push necessary state changes back into the socket
1342  * buffer so that other consumers see the values consistently.  'nextrecord'
1343  * is the callers locally stored value of the original value of
1344  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1345  * NOTE: 'nextrecord' may be NULL.
1346  */
1347 static __inline void
1348 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1349 {
1350 
1351 	SOCKBUF_LOCK_ASSERT(sb);
1352 	/*
1353 	 * First, update for the new value of nextrecord.  If necessary, make
1354 	 * it the first record.
1355 	 */
1356 	if (sb->sb_mb != NULL)
1357 		sb->sb_mb->m_nextpkt = nextrecord;
1358 	else
1359 		sb->sb_mb = nextrecord;
1360 
1361         /*
1362          * Now update any dependent socket buffer fields to reflect the new
1363          * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
1364 	 * addition of a second clause that takes care of the case where
1365 	 * sb_mb has been updated, but remains the last record.
1366          */
1367         if (sb->sb_mb == NULL) {
1368                 sb->sb_mbtail = NULL;
1369                 sb->sb_lastrecord = NULL;
1370         } else if (sb->sb_mb->m_nextpkt == NULL)
1371                 sb->sb_lastrecord = sb->sb_mb;
1372 }
1373 
1374 
1375 /*
1376  * Implement receive operations on a socket.  We depend on the way that
1377  * records are added to the sockbuf by sbappend.  In particular, each record
1378  * (mbufs linked through m_next) must begin with an address if the protocol
1379  * so specifies, followed by an optional mbuf or mbufs containing ancillary
1380  * data, and then zero or more mbufs of data.  In order to allow parallelism
1381  * between network receive and copying to user space, as well as avoid
1382  * sleeping with a mutex held, we release the socket buffer mutex during the
1383  * user space copy.  Although the sockbuf is locked, new data may still be
1384  * appended, and thus we must maintain consistency of the sockbuf during that
1385  * time.
1386  *
1387  * The caller may receive the data as a single mbuf chain by supplying an
1388  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
1389  * the count in uio_resid.
1390  */
1391 int
1392 soreceive_generic(so, psa, uio, mp0, controlp, flagsp)
1393 	struct socket *so;
1394 	struct sockaddr **psa;
1395 	struct uio *uio;
1396 	struct mbuf **mp0;
1397 	struct mbuf **controlp;
1398 	int *flagsp;
1399 {
1400 	struct mbuf *m, **mp;
1401 	int flags, len, error, offset;
1402 	struct protosw *pr = so->so_proto;
1403 	struct mbuf *nextrecord;
1404 	int moff, type = 0;
1405 	int orig_resid = uio->uio_resid;
1406 
1407 	mp = mp0;
1408 	if (psa != NULL)
1409 		*psa = NULL;
1410 	if (controlp != NULL)
1411 		*controlp = NULL;
1412 	if (flagsp != NULL)
1413 		flags = *flagsp &~ MSG_EOR;
1414 	else
1415 		flags = 0;
1416 	if (flags & MSG_OOB)
1417 		return (soreceive_rcvoob(so, uio, flags));
1418 	if (mp != NULL)
1419 		*mp = NULL;
1420 	if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1421 	    && uio->uio_resid)
1422 		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
1423 
1424 	SOCKBUF_LOCK(&so->so_rcv);
1425 restart:
1426 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1427 	error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1428 	if (error)
1429 		goto out;
1430 
1431 	m = so->so_rcv.sb_mb;
1432 	/*
1433 	 * If we have less data than requested, block awaiting more (subject
1434 	 * to any timeout) if:
1435 	 *   1. the current count is less than the low water mark, or
1436 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1437 	 *	receive operation at once if we block (resid <= hiwat).
1438 	 *   3. MSG_DONTWAIT is not set
1439 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1440 	 * we have to do the receive in sections, and thus risk returning a
1441 	 * short count if a timeout or signal occurs after we start.
1442 	 */
1443 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1444 	    so->so_rcv.sb_cc < uio->uio_resid) &&
1445 	    (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1446 	    ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1447 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1448 		KASSERT(m != NULL || !so->so_rcv.sb_cc,
1449 		    ("receive: m == %p so->so_rcv.sb_cc == %u",
1450 		    m, so->so_rcv.sb_cc));
1451 		if (so->so_error) {
1452 			if (m != NULL)
1453 				goto dontblock;
1454 			error = so->so_error;
1455 			if ((flags & MSG_PEEK) == 0)
1456 				so->so_error = 0;
1457 			goto release;
1458 		}
1459 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1460 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1461 			if (m)
1462 				goto dontblock;
1463 			else
1464 				goto release;
1465 		}
1466 		for (; m != NULL; m = m->m_next)
1467 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1468 				m = so->so_rcv.sb_mb;
1469 				goto dontblock;
1470 			}
1471 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1472 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1473 			error = ENOTCONN;
1474 			goto release;
1475 		}
1476 		if (uio->uio_resid == 0)
1477 			goto release;
1478 		if ((so->so_state & SS_NBIO) ||
1479 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1480 			error = EWOULDBLOCK;
1481 			goto release;
1482 		}
1483 		SBLASTRECORDCHK(&so->so_rcv);
1484 		SBLASTMBUFCHK(&so->so_rcv);
1485 		sbunlock(&so->so_rcv);
1486 		error = sbwait(&so->so_rcv);
1487 		if (error)
1488 			goto out;
1489 		goto restart;
1490 	}
1491 dontblock:
1492 	/*
1493 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1494 	 * pointer to the next record in the socket buffer.  We must keep the
1495 	 * various socket buffer pointers and local stack versions of the
1496 	 * pointers in sync, pushing out modifications before dropping the
1497 	 * socket buffer mutex, and re-reading them when picking it up.
1498 	 *
1499 	 * Otherwise, we will race with the network stack appending new data
1500 	 * or records onto the socket buffer by using inconsistent/stale
1501 	 * versions of the field, possibly resulting in socket buffer
1502 	 * corruption.
1503 	 *
1504 	 * By holding the high-level sblock(), we prevent simultaneous
1505 	 * readers from pulling off the front of the socket buffer.
1506 	 */
1507 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1508 	if (uio->uio_td)
1509 		uio->uio_td->td_proc->p_stats->p_ru.ru_msgrcv++;
1510 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1511 	SBLASTRECORDCHK(&so->so_rcv);
1512 	SBLASTMBUFCHK(&so->so_rcv);
1513 	nextrecord = m->m_nextpkt;
1514 	if (pr->pr_flags & PR_ADDR) {
1515 		KASSERT(m->m_type == MT_SONAME,
1516 		    ("m->m_type == %d", m->m_type));
1517 		orig_resid = 0;
1518 		if (psa != NULL)
1519 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
1520 			    M_NOWAIT);
1521 		if (flags & MSG_PEEK) {
1522 			m = m->m_next;
1523 		} else {
1524 			sbfree(&so->so_rcv, m);
1525 			so->so_rcv.sb_mb = m_free(m);
1526 			m = so->so_rcv.sb_mb;
1527 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1528 		}
1529 	}
1530 
1531 	/*
1532 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1533 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1534 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1535 	 * perform externalization (or freeing if controlp == NULL).
1536 	 */
1537 	if (m != NULL && m->m_type == MT_CONTROL) {
1538 		struct mbuf *cm = NULL, *cmn;
1539 		struct mbuf **cme = &cm;
1540 
1541 		do {
1542 			if (flags & MSG_PEEK) {
1543 				if (controlp != NULL) {
1544 					*controlp = m_copy(m, 0, m->m_len);
1545 					controlp = &(*controlp)->m_next;
1546 				}
1547 				m = m->m_next;
1548 			} else {
1549 				sbfree(&so->so_rcv, m);
1550 				so->so_rcv.sb_mb = m->m_next;
1551 				m->m_next = NULL;
1552 				*cme = m;
1553 				cme = &(*cme)->m_next;
1554 				m = so->so_rcv.sb_mb;
1555 			}
1556 		} while (m != NULL && m->m_type == MT_CONTROL);
1557 		if ((flags & MSG_PEEK) == 0)
1558 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1559 		while (cm != NULL) {
1560 			cmn = cm->m_next;
1561 			cm->m_next = NULL;
1562 			if (pr->pr_domain->dom_externalize != NULL) {
1563 				SOCKBUF_UNLOCK(&so->so_rcv);
1564 				error = (*pr->pr_domain->dom_externalize)
1565 				    (cm, controlp);
1566 				SOCKBUF_LOCK(&so->so_rcv);
1567 			} else if (controlp != NULL)
1568 				*controlp = cm;
1569 			else
1570 				m_freem(cm);
1571 			if (controlp != NULL) {
1572 				orig_resid = 0;
1573 				while (*controlp != NULL)
1574 					controlp = &(*controlp)->m_next;
1575 			}
1576 			cm = cmn;
1577 		}
1578 		if (so->so_rcv.sb_mb)
1579 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1580 		else
1581 			nextrecord = NULL;
1582 		orig_resid = 0;
1583 	}
1584 	if (m != NULL) {
1585 		if ((flags & MSG_PEEK) == 0) {
1586 			KASSERT(m->m_nextpkt == nextrecord,
1587 			    ("soreceive: post-control, nextrecord !sync"));
1588 			if (nextrecord == NULL) {
1589 				KASSERT(so->so_rcv.sb_mb == m,
1590 				    ("soreceive: post-control, sb_mb!=m"));
1591 				KASSERT(so->so_rcv.sb_lastrecord == m,
1592 				    ("soreceive: post-control, lastrecord!=m"));
1593 			}
1594 		}
1595 		type = m->m_type;
1596 		if (type == MT_OOBDATA)
1597 			flags |= MSG_OOB;
1598 	} else {
1599 		if ((flags & MSG_PEEK) == 0) {
1600 			KASSERT(so->so_rcv.sb_mb == nextrecord,
1601 			    ("soreceive: sb_mb != nextrecord"));
1602 			if (so->so_rcv.sb_mb == NULL) {
1603 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
1604 				    ("soreceive: sb_lastercord != NULL"));
1605 			}
1606 		}
1607 	}
1608 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1609 	SBLASTRECORDCHK(&so->so_rcv);
1610 	SBLASTMBUFCHK(&so->so_rcv);
1611 
1612 	/*
1613 	 * Now continue to read any data mbufs off of the head of the socket
1614 	 * buffer until the read request is satisfied.  Note that 'type' is
1615 	 * used to store the type of any mbuf reads that have happened so far
1616 	 * such that soreceive() can stop reading if the type changes, which
1617 	 * causes soreceive() to return only one of regular data and inline
1618 	 * out-of-band data in a single socket receive operation.
1619 	 */
1620 	moff = 0;
1621 	offset = 0;
1622 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1623 		/*
1624 		 * If the type of mbuf has changed since the last mbuf
1625 		 * examined ('type'), end the receive operation.
1626 	 	 */
1627 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1628 		if (m->m_type == MT_OOBDATA) {
1629 			if (type != MT_OOBDATA)
1630 				break;
1631 		} else if (type == MT_OOBDATA)
1632 			break;
1633 		else
1634 		    KASSERT(m->m_type == MT_DATA,
1635 			("m->m_type == %d", m->m_type));
1636 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1637 		len = uio->uio_resid;
1638 		if (so->so_oobmark && len > so->so_oobmark - offset)
1639 			len = so->so_oobmark - offset;
1640 		if (len > m->m_len - moff)
1641 			len = m->m_len - moff;
1642 		/*
1643 		 * If mp is set, just pass back the mbufs.  Otherwise copy
1644 		 * them out via the uio, then free.  Sockbuf must be
1645 		 * consistent here (points to current mbuf, it points to next
1646 		 * record) when we drop priority; we must note any additions
1647 		 * to the sockbuf when we block interrupts again.
1648 		 */
1649 		if (mp == NULL) {
1650 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1651 			SBLASTRECORDCHK(&so->so_rcv);
1652 			SBLASTMBUFCHK(&so->so_rcv);
1653 			SOCKBUF_UNLOCK(&so->so_rcv);
1654 #ifdef ZERO_COPY_SOCKETS
1655 			if (so_zero_copy_receive) {
1656 				int disposable;
1657 
1658 				if ((m->m_flags & M_EXT)
1659 				 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1660 					disposable = 1;
1661 				else
1662 					disposable = 0;
1663 
1664 				error = uiomoveco(mtod(m, char *) + moff,
1665 						  (int)len, uio,
1666 						  disposable);
1667 			} else
1668 #endif /* ZERO_COPY_SOCKETS */
1669 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1670 			SOCKBUF_LOCK(&so->so_rcv);
1671 			if (error)
1672 				goto release;
1673 		} else
1674 			uio->uio_resid -= len;
1675 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1676 		if (len == m->m_len - moff) {
1677 			if (m->m_flags & M_EOR)
1678 				flags |= MSG_EOR;
1679 			if (flags & MSG_PEEK) {
1680 				m = m->m_next;
1681 				moff = 0;
1682 			} else {
1683 				nextrecord = m->m_nextpkt;
1684 				sbfree(&so->so_rcv, m);
1685 				if (mp != NULL) {
1686 					*mp = m;
1687 					mp = &m->m_next;
1688 					so->so_rcv.sb_mb = m = m->m_next;
1689 					*mp = NULL;
1690 				} else {
1691 					so->so_rcv.sb_mb = m_free(m);
1692 					m = so->so_rcv.sb_mb;
1693 				}
1694 				sockbuf_pushsync(&so->so_rcv, nextrecord);
1695 				SBLASTRECORDCHK(&so->so_rcv);
1696 				SBLASTMBUFCHK(&so->so_rcv);
1697 			}
1698 		} else {
1699 			if (flags & MSG_PEEK)
1700 				moff += len;
1701 			else {
1702 				if (mp != NULL) {
1703 					int copy_flag;
1704 
1705 					if (flags & MSG_DONTWAIT)
1706 						copy_flag = M_DONTWAIT;
1707 					else
1708 						copy_flag = M_TRYWAIT;
1709 					if (copy_flag == M_TRYWAIT)
1710 						SOCKBUF_UNLOCK(&so->so_rcv);
1711 					*mp = m_copym(m, 0, len, copy_flag);
1712 					if (copy_flag == M_TRYWAIT)
1713 						SOCKBUF_LOCK(&so->so_rcv);
1714  					if (*mp == NULL) {
1715  						/*
1716  						 * m_copym() couldn't
1717 						 * allocate an mbuf.  Adjust
1718 						 * uio_resid back (it was
1719 						 * adjusted down by len
1720 						 * bytes, which we didn't end
1721 						 * up "copying" over).
1722  						 */
1723  						uio->uio_resid += len;
1724  						break;
1725  					}
1726 				}
1727 				m->m_data += len;
1728 				m->m_len -= len;
1729 				so->so_rcv.sb_cc -= len;
1730 			}
1731 		}
1732 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1733 		if (so->so_oobmark) {
1734 			if ((flags & MSG_PEEK) == 0) {
1735 				so->so_oobmark -= len;
1736 				if (so->so_oobmark == 0) {
1737 					so->so_rcv.sb_state |= SBS_RCVATMARK;
1738 					break;
1739 				}
1740 			} else {
1741 				offset += len;
1742 				if (offset == so->so_oobmark)
1743 					break;
1744 			}
1745 		}
1746 		if (flags & MSG_EOR)
1747 			break;
1748 		/*
1749 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1750 		 * must not quit until "uio->uio_resid == 0" or an error
1751 		 * termination.  If a signal/timeout occurs, return with a
1752 		 * short count but without error.  Keep sockbuf locked
1753 		 * against other readers.
1754 		 */
1755 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1756 		    !sosendallatonce(so) && nextrecord == NULL) {
1757 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1758 			if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE)
1759 				break;
1760 			/*
1761 			 * Notify the protocol that some data has been
1762 			 * drained before blocking.
1763 			 */
1764 			if (pr->pr_flags & PR_WANTRCVD) {
1765 				SOCKBUF_UNLOCK(&so->so_rcv);
1766 				(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1767 				SOCKBUF_LOCK(&so->so_rcv);
1768 			}
1769 			SBLASTRECORDCHK(&so->so_rcv);
1770 			SBLASTMBUFCHK(&so->so_rcv);
1771 			error = sbwait(&so->so_rcv);
1772 			if (error)
1773 				goto release;
1774 			m = so->so_rcv.sb_mb;
1775 			if (m != NULL)
1776 				nextrecord = m->m_nextpkt;
1777 		}
1778 	}
1779 
1780 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1781 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1782 		flags |= MSG_TRUNC;
1783 		if ((flags & MSG_PEEK) == 0)
1784 			(void) sbdroprecord_locked(&so->so_rcv);
1785 	}
1786 	if ((flags & MSG_PEEK) == 0) {
1787 		if (m == NULL) {
1788 			/*
1789 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1790 			 * part makes sure sb_lastrecord is up-to-date if
1791 			 * there is still data in the socket buffer.
1792 			 */
1793 			so->so_rcv.sb_mb = nextrecord;
1794 			if (so->so_rcv.sb_mb == NULL) {
1795 				so->so_rcv.sb_mbtail = NULL;
1796 				so->so_rcv.sb_lastrecord = NULL;
1797 			} else if (nextrecord->m_nextpkt == NULL)
1798 				so->so_rcv.sb_lastrecord = nextrecord;
1799 		}
1800 		SBLASTRECORDCHK(&so->so_rcv);
1801 		SBLASTMBUFCHK(&so->so_rcv);
1802 		/*
1803 		 * If soreceive() is being done from the socket callback,
1804 		 * then don't need to generate ACK to peer to update window,
1805 		 * since ACK will be generated on return to TCP.
1806 		 */
1807 		if (!(flags & MSG_SOCALLBCK) &&
1808 		    (pr->pr_flags & PR_WANTRCVD)) {
1809 			SOCKBUF_UNLOCK(&so->so_rcv);
1810 			(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1811 			SOCKBUF_LOCK(&so->so_rcv);
1812 		}
1813 	}
1814 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1815 	if (orig_resid == uio->uio_resid && orig_resid &&
1816 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1817 		sbunlock(&so->so_rcv);
1818 		goto restart;
1819 	}
1820 
1821 	if (flagsp != NULL)
1822 		*flagsp |= flags;
1823 release:
1824 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1825 	sbunlock(&so->so_rcv);
1826 out:
1827 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1828 	SOCKBUF_UNLOCK(&so->so_rcv);
1829 	return (error);
1830 }
1831 
1832 int
1833 soreceive(so, psa, uio, mp0, controlp, flagsp)
1834 	struct socket *so;
1835 	struct sockaddr **psa;
1836 	struct uio *uio;
1837 	struct mbuf **mp0;
1838 	struct mbuf **controlp;
1839 	int *flagsp;
1840 {
1841 
1842 	/* XXXRW: Temporary debugging. */
1843 	KASSERT(so->so_proto->pr_usrreqs->pru_soreceive != soreceive,
1844 	    ("soreceive: protocol calls soreceive"));
1845 
1846 	return (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
1847 	    controlp, flagsp));
1848 }
1849 
1850 int
1851 soshutdown(so, how)
1852 	struct socket *so;
1853 	int how;
1854 {
1855 	struct protosw *pr = so->so_proto;
1856 
1857 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1858 		return (EINVAL);
1859 
1860 	if (how != SHUT_WR)
1861 		sorflush(so);
1862 	if (how != SHUT_RD)
1863 		return ((*pr->pr_usrreqs->pru_shutdown)(so));
1864 	return (0);
1865 }
1866 
1867 void
1868 sorflush(so)
1869 	struct socket *so;
1870 {
1871 	struct sockbuf *sb = &so->so_rcv;
1872 	struct protosw *pr = so->so_proto;
1873 	struct sockbuf asb;
1874 
1875 	/*
1876 	 * XXXRW: This is quite ugly.  Previously, this code made a copy of
1877 	 * the socket buffer, then zero'd the original to clear the buffer
1878 	 * fields.  However, with mutexes in the socket buffer, this causes
1879 	 * problems.  We only clear the zeroable bits of the original;
1880 	 * however, we have to initialize and destroy the mutex in the copy
1881 	 * so that dom_dispose() and sbrelease() can lock t as needed.
1882 	 */
1883 	SOCKBUF_LOCK(sb);
1884 	sb->sb_flags |= SB_NOINTR;
1885 	(void) sblock(sb, M_WAITOK);
1886 	/*
1887 	 * socantrcvmore_locked() drops the socket buffer mutex so that it
1888 	 * can safely perform wakeups.  Re-acquire the mutex before
1889 	 * continuing.
1890 	 */
1891 	socantrcvmore_locked(so);
1892 	SOCKBUF_LOCK(sb);
1893 	sbunlock(sb);
1894 	/*
1895 	 * Invalidate/clear most of the sockbuf structure, but leave selinfo
1896 	 * and mutex data unchanged.
1897 	 */
1898 	bzero(&asb, offsetof(struct sockbuf, sb_startzero));
1899 	bcopy(&sb->sb_startzero, &asb.sb_startzero,
1900 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1901 	bzero(&sb->sb_startzero,
1902 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1903 	SOCKBUF_UNLOCK(sb);
1904 
1905 	SOCKBUF_LOCK_INIT(&asb, "so_rcv");
1906 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1907 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1908 	sbrelease(&asb, so);
1909 	SOCKBUF_LOCK_DESTROY(&asb);
1910 }
1911 
1912 /*
1913  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
1914  * additional variant to handle the case where the option value needs to be
1915  * some kind of integer, but not a specific size.  In addition to their use
1916  * here, these functions are also called by the protocol-level pr_ctloutput()
1917  * routines.
1918  */
1919 int
1920 sooptcopyin(sopt, buf, len, minlen)
1921 	struct	sockopt *sopt;
1922 	void	*buf;
1923 	size_t	len;
1924 	size_t	minlen;
1925 {
1926 	size_t	valsize;
1927 
1928 	/*
1929 	 * If the user gives us more than we wanted, we ignore it, but if we
1930 	 * don't get the minimum length the caller wants, we return EINVAL.
1931 	 * On success, sopt->sopt_valsize is set to however much we actually
1932 	 * retrieved.
1933 	 */
1934 	if ((valsize = sopt->sopt_valsize) < minlen)
1935 		return EINVAL;
1936 	if (valsize > len)
1937 		sopt->sopt_valsize = valsize = len;
1938 
1939 	if (sopt->sopt_td != NULL)
1940 		return (copyin(sopt->sopt_val, buf, valsize));
1941 
1942 	bcopy(sopt->sopt_val, buf, valsize);
1943 	return (0);
1944 }
1945 
1946 /*
1947  * Kernel version of setsockopt(2).
1948  *
1949  * XXX: optlen is size_t, not socklen_t
1950  */
1951 int
1952 so_setsockopt(struct socket *so, int level, int optname, void *optval,
1953     size_t optlen)
1954 {
1955 	struct sockopt sopt;
1956 
1957 	sopt.sopt_level = level;
1958 	sopt.sopt_name = optname;
1959 	sopt.sopt_dir = SOPT_SET;
1960 	sopt.sopt_val = optval;
1961 	sopt.sopt_valsize = optlen;
1962 	sopt.sopt_td = NULL;
1963 	return (sosetopt(so, &sopt));
1964 }
1965 
1966 int
1967 sosetopt(so, sopt)
1968 	struct socket *so;
1969 	struct sockopt *sopt;
1970 {
1971 	int	error, optval;
1972 	struct	linger l;
1973 	struct	timeval tv;
1974 	u_long  val;
1975 #ifdef MAC
1976 	struct mac extmac;
1977 #endif
1978 
1979 	error = 0;
1980 	if (sopt->sopt_level != SOL_SOCKET) {
1981 		if (so->so_proto && so->so_proto->pr_ctloutput)
1982 			return ((*so->so_proto->pr_ctloutput)
1983 				  (so, sopt));
1984 		error = ENOPROTOOPT;
1985 	} else {
1986 		switch (sopt->sopt_name) {
1987 #ifdef INET
1988 		case SO_ACCEPTFILTER:
1989 			error = do_setopt_accept_filter(so, sopt);
1990 			if (error)
1991 				goto bad;
1992 			break;
1993 #endif
1994 		case SO_LINGER:
1995 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1996 			if (error)
1997 				goto bad;
1998 
1999 			SOCK_LOCK(so);
2000 			so->so_linger = l.l_linger;
2001 			if (l.l_onoff)
2002 				so->so_options |= SO_LINGER;
2003 			else
2004 				so->so_options &= ~SO_LINGER;
2005 			SOCK_UNLOCK(so);
2006 			break;
2007 
2008 		case SO_DEBUG:
2009 		case SO_KEEPALIVE:
2010 		case SO_DONTROUTE:
2011 		case SO_USELOOPBACK:
2012 		case SO_BROADCAST:
2013 		case SO_REUSEADDR:
2014 		case SO_REUSEPORT:
2015 		case SO_OOBINLINE:
2016 		case SO_TIMESTAMP:
2017 		case SO_BINTIME:
2018 		case SO_NOSIGPIPE:
2019 			error = sooptcopyin(sopt, &optval, sizeof optval,
2020 					    sizeof optval);
2021 			if (error)
2022 				goto bad;
2023 			SOCK_LOCK(so);
2024 			if (optval)
2025 				so->so_options |= sopt->sopt_name;
2026 			else
2027 				so->so_options &= ~sopt->sopt_name;
2028 			SOCK_UNLOCK(so);
2029 			break;
2030 
2031 		case SO_SNDBUF:
2032 		case SO_RCVBUF:
2033 		case SO_SNDLOWAT:
2034 		case SO_RCVLOWAT:
2035 			error = sooptcopyin(sopt, &optval, sizeof optval,
2036 					    sizeof optval);
2037 			if (error)
2038 				goto bad;
2039 
2040 			/*
2041 			 * Values < 1 make no sense for any of these options,
2042 			 * so disallow them.
2043 			 */
2044 			if (optval < 1) {
2045 				error = EINVAL;
2046 				goto bad;
2047 			}
2048 
2049 			switch (sopt->sopt_name) {
2050 			case SO_SNDBUF:
2051 			case SO_RCVBUF:
2052 				if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2053 				    &so->so_snd : &so->so_rcv, (u_long)optval,
2054 				    so, curthread) == 0) {
2055 					error = ENOBUFS;
2056 					goto bad;
2057 				}
2058 				break;
2059 
2060 			/*
2061 			 * Make sure the low-water is never greater than the
2062 			 * high-water.
2063 			 */
2064 			case SO_SNDLOWAT:
2065 				SOCKBUF_LOCK(&so->so_snd);
2066 				so->so_snd.sb_lowat =
2067 				    (optval > so->so_snd.sb_hiwat) ?
2068 				    so->so_snd.sb_hiwat : optval;
2069 				SOCKBUF_UNLOCK(&so->so_snd);
2070 				break;
2071 			case SO_RCVLOWAT:
2072 				SOCKBUF_LOCK(&so->so_rcv);
2073 				so->so_rcv.sb_lowat =
2074 				    (optval > so->so_rcv.sb_hiwat) ?
2075 				    so->so_rcv.sb_hiwat : optval;
2076 				SOCKBUF_UNLOCK(&so->so_rcv);
2077 				break;
2078 			}
2079 			break;
2080 
2081 		case SO_SNDTIMEO:
2082 		case SO_RCVTIMEO:
2083 #ifdef COMPAT_IA32
2084 			if (curthread->td_proc->p_sysent == &ia32_freebsd_sysvec) {
2085 				struct timeval32 tv32;
2086 
2087 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
2088 				    sizeof tv32);
2089 				CP(tv32, tv, tv_sec);
2090 				CP(tv32, tv, tv_usec);
2091 			} else
2092 #endif
2093 				error = sooptcopyin(sopt, &tv, sizeof tv,
2094 				    sizeof tv);
2095 			if (error)
2096 				goto bad;
2097 
2098 			/* assert(hz > 0); */
2099 			if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2100 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2101 				error = EDOM;
2102 				goto bad;
2103 			}
2104 			/* assert(tick > 0); */
2105 			/* assert(ULONG_MAX - INT_MAX >= 1000000); */
2106 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
2107 			if (val > INT_MAX) {
2108 				error = EDOM;
2109 				goto bad;
2110 			}
2111 			if (val == 0 && tv.tv_usec != 0)
2112 				val = 1;
2113 
2114 			switch (sopt->sopt_name) {
2115 			case SO_SNDTIMEO:
2116 				so->so_snd.sb_timeo = val;
2117 				break;
2118 			case SO_RCVTIMEO:
2119 				so->so_rcv.sb_timeo = val;
2120 				break;
2121 			}
2122 			break;
2123 
2124 		case SO_LABEL:
2125 #ifdef MAC
2126 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
2127 			    sizeof extmac);
2128 			if (error)
2129 				goto bad;
2130 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2131 			    so, &extmac);
2132 #else
2133 			error = EOPNOTSUPP;
2134 #endif
2135 			break;
2136 
2137 		default:
2138 			error = ENOPROTOOPT;
2139 			break;
2140 		}
2141 		if (error == 0 && so->so_proto != NULL &&
2142 		    so->so_proto->pr_ctloutput != NULL) {
2143 			(void) ((*so->so_proto->pr_ctloutput)
2144 				  (so, sopt));
2145 		}
2146 	}
2147 bad:
2148 	return (error);
2149 }
2150 
2151 /*
2152  * Helper routine for getsockopt.
2153  */
2154 int
2155 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2156 {
2157 	int	error;
2158 	size_t	valsize;
2159 
2160 	error = 0;
2161 
2162 	/*
2163 	 * Documented get behavior is that we always return a value, possibly
2164 	 * truncated to fit in the user's buffer.  Traditional behavior is
2165 	 * that we always tell the user precisely how much we copied, rather
2166 	 * than something useful like the total amount we had available for
2167 	 * her.  Note that this interface is not idempotent; the entire
2168 	 * answer must generated ahead of time.
2169 	 */
2170 	valsize = min(len, sopt->sopt_valsize);
2171 	sopt->sopt_valsize = valsize;
2172 	if (sopt->sopt_val != NULL) {
2173 		if (sopt->sopt_td != NULL)
2174 			error = copyout(buf, sopt->sopt_val, valsize);
2175 		else
2176 			bcopy(buf, sopt->sopt_val, valsize);
2177 	}
2178 	return (error);
2179 }
2180 
2181 int
2182 sogetopt(so, sopt)
2183 	struct socket *so;
2184 	struct sockopt *sopt;
2185 {
2186 	int	error, optval;
2187 	struct	linger l;
2188 	struct	timeval tv;
2189 #ifdef MAC
2190 	struct mac extmac;
2191 #endif
2192 
2193 	error = 0;
2194 	if (sopt->sopt_level != SOL_SOCKET) {
2195 		if (so->so_proto && so->so_proto->pr_ctloutput) {
2196 			return ((*so->so_proto->pr_ctloutput)
2197 				  (so, sopt));
2198 		} else
2199 			return (ENOPROTOOPT);
2200 	} else {
2201 		switch (sopt->sopt_name) {
2202 #ifdef INET
2203 		case SO_ACCEPTFILTER:
2204 			error = do_getopt_accept_filter(so, sopt);
2205 			break;
2206 #endif
2207 		case SO_LINGER:
2208 			SOCK_LOCK(so);
2209 			l.l_onoff = so->so_options & SO_LINGER;
2210 			l.l_linger = so->so_linger;
2211 			SOCK_UNLOCK(so);
2212 			error = sooptcopyout(sopt, &l, sizeof l);
2213 			break;
2214 
2215 		case SO_USELOOPBACK:
2216 		case SO_DONTROUTE:
2217 		case SO_DEBUG:
2218 		case SO_KEEPALIVE:
2219 		case SO_REUSEADDR:
2220 		case SO_REUSEPORT:
2221 		case SO_BROADCAST:
2222 		case SO_OOBINLINE:
2223 		case SO_ACCEPTCONN:
2224 		case SO_TIMESTAMP:
2225 		case SO_BINTIME:
2226 		case SO_NOSIGPIPE:
2227 			optval = so->so_options & sopt->sopt_name;
2228 integer:
2229 			error = sooptcopyout(sopt, &optval, sizeof optval);
2230 			break;
2231 
2232 		case SO_TYPE:
2233 			optval = so->so_type;
2234 			goto integer;
2235 
2236 		case SO_ERROR:
2237 			SOCK_LOCK(so);
2238 			optval = so->so_error;
2239 			so->so_error = 0;
2240 			SOCK_UNLOCK(so);
2241 			goto integer;
2242 
2243 		case SO_SNDBUF:
2244 			optval = so->so_snd.sb_hiwat;
2245 			goto integer;
2246 
2247 		case SO_RCVBUF:
2248 			optval = so->so_rcv.sb_hiwat;
2249 			goto integer;
2250 
2251 		case SO_SNDLOWAT:
2252 			optval = so->so_snd.sb_lowat;
2253 			goto integer;
2254 
2255 		case SO_RCVLOWAT:
2256 			optval = so->so_rcv.sb_lowat;
2257 			goto integer;
2258 
2259 		case SO_SNDTIMEO:
2260 		case SO_RCVTIMEO:
2261 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
2262 				  so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2263 
2264 			tv.tv_sec = optval / hz;
2265 			tv.tv_usec = (optval % hz) * tick;
2266 #ifdef COMPAT_IA32
2267 			if (curthread->td_proc->p_sysent == &ia32_freebsd_sysvec) {
2268 				struct timeval32 tv32;
2269 
2270 				CP(tv, tv32, tv_sec);
2271 				CP(tv, tv32, tv_usec);
2272 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
2273 			} else
2274 #endif
2275 				error = sooptcopyout(sopt, &tv, sizeof tv);
2276 			break;
2277 
2278 		case SO_LABEL:
2279 #ifdef MAC
2280 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2281 			    sizeof(extmac));
2282 			if (error)
2283 				return (error);
2284 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2285 			    so, &extmac);
2286 			if (error)
2287 				return (error);
2288 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2289 #else
2290 			error = EOPNOTSUPP;
2291 #endif
2292 			break;
2293 
2294 		case SO_PEERLABEL:
2295 #ifdef MAC
2296 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2297 			    sizeof(extmac));
2298 			if (error)
2299 				return (error);
2300 			error = mac_getsockopt_peerlabel(
2301 			    sopt->sopt_td->td_ucred, so, &extmac);
2302 			if (error)
2303 				return (error);
2304 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2305 #else
2306 			error = EOPNOTSUPP;
2307 #endif
2308 			break;
2309 
2310 		case SO_LISTENQLIMIT:
2311 			optval = so->so_qlimit;
2312 			goto integer;
2313 
2314 		case SO_LISTENQLEN:
2315 			optval = so->so_qlen;
2316 			goto integer;
2317 
2318 		case SO_LISTENINCQLEN:
2319 			optval = so->so_incqlen;
2320 			goto integer;
2321 
2322 		default:
2323 			error = ENOPROTOOPT;
2324 			break;
2325 		}
2326 		return (error);
2327 	}
2328 }
2329 
2330 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2331 int
2332 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2333 {
2334 	struct mbuf *m, *m_prev;
2335 	int sopt_size = sopt->sopt_valsize;
2336 
2337 	MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA);
2338 	if (m == NULL)
2339 		return ENOBUFS;
2340 	if (sopt_size > MLEN) {
2341 		MCLGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT);
2342 		if ((m->m_flags & M_EXT) == 0) {
2343 			m_free(m);
2344 			return ENOBUFS;
2345 		}
2346 		m->m_len = min(MCLBYTES, sopt_size);
2347 	} else {
2348 		m->m_len = min(MLEN, sopt_size);
2349 	}
2350 	sopt_size -= m->m_len;
2351 	*mp = m;
2352 	m_prev = m;
2353 
2354 	while (sopt_size) {
2355 		MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA);
2356 		if (m == NULL) {
2357 			m_freem(*mp);
2358 			return ENOBUFS;
2359 		}
2360 		if (sopt_size > MLEN) {
2361 			MCLGET(m, sopt->sopt_td != NULL ? M_TRYWAIT :
2362 			    M_DONTWAIT);
2363 			if ((m->m_flags & M_EXT) == 0) {
2364 				m_freem(m);
2365 				m_freem(*mp);
2366 				return ENOBUFS;
2367 			}
2368 			m->m_len = min(MCLBYTES, sopt_size);
2369 		} else {
2370 			m->m_len = min(MLEN, sopt_size);
2371 		}
2372 		sopt_size -= m->m_len;
2373 		m_prev->m_next = m;
2374 		m_prev = m;
2375 	}
2376 	return (0);
2377 }
2378 
2379 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2380 int
2381 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2382 {
2383 	struct mbuf *m0 = m;
2384 
2385 	if (sopt->sopt_val == NULL)
2386 		return (0);
2387 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2388 		if (sopt->sopt_td != NULL) {
2389 			int error;
2390 
2391 			error = copyin(sopt->sopt_val, mtod(m, char *),
2392 				       m->m_len);
2393 			if (error != 0) {
2394 				m_freem(m0);
2395 				return(error);
2396 			}
2397 		} else
2398 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2399 		sopt->sopt_valsize -= m->m_len;
2400 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2401 		m = m->m_next;
2402 	}
2403 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2404 		panic("ip6_sooptmcopyin");
2405 	return (0);
2406 }
2407 
2408 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2409 int
2410 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2411 {
2412 	struct mbuf *m0 = m;
2413 	size_t valsize = 0;
2414 
2415 	if (sopt->sopt_val == NULL)
2416 		return (0);
2417 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2418 		if (sopt->sopt_td != NULL) {
2419 			int error;
2420 
2421 			error = copyout(mtod(m, char *), sopt->sopt_val,
2422 				       m->m_len);
2423 			if (error != 0) {
2424 				m_freem(m0);
2425 				return(error);
2426 			}
2427 		} else
2428 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2429 	       sopt->sopt_valsize -= m->m_len;
2430 	       sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2431 	       valsize += m->m_len;
2432 	       m = m->m_next;
2433 	}
2434 	if (m != NULL) {
2435 		/* enough soopt buffer should be given from user-land */
2436 		m_freem(m0);
2437 		return(EINVAL);
2438 	}
2439 	sopt->sopt_valsize = valsize;
2440 	return (0);
2441 }
2442 
2443 /*
2444  * sohasoutofband(): protocol notifies socket layer of the arrival of new
2445  * out-of-band data, which will then notify socket consumers.
2446  */
2447 void
2448 sohasoutofband(so)
2449 	struct socket *so;
2450 {
2451 	if (so->so_sigio != NULL)
2452 		pgsigio(&so->so_sigio, SIGURG, 0);
2453 	selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2454 }
2455 
2456 int
2457 sopoll(struct socket *so, int events, struct ucred *active_cred,
2458     struct thread *td)
2459 {
2460 
2461 	/* XXXRW: Temporary debugging. */
2462 	KASSERT(so->so_proto->pr_usrreqs->pru_sopoll != sopoll,
2463 	    ("sopoll: protocol calls sopoll"));
2464 
2465 	return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
2466 	    td));
2467 }
2468 
2469 int
2470 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
2471     struct thread *td)
2472 {
2473 	int revents = 0;
2474 
2475 	SOCKBUF_LOCK(&so->so_snd);
2476 	SOCKBUF_LOCK(&so->so_rcv);
2477 	if (events & (POLLIN | POLLRDNORM))
2478 		if (soreadable(so))
2479 			revents |= events & (POLLIN | POLLRDNORM);
2480 
2481 	if (events & POLLINIGNEOF)
2482 		if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat ||
2483 		    !TAILQ_EMPTY(&so->so_comp) || so->so_error)
2484 			revents |= POLLINIGNEOF;
2485 
2486 	if (events & (POLLOUT | POLLWRNORM))
2487 		if (sowriteable(so))
2488 			revents |= events & (POLLOUT | POLLWRNORM);
2489 
2490 	if (events & (POLLPRI | POLLRDBAND))
2491 		if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2492 			revents |= events & (POLLPRI | POLLRDBAND);
2493 
2494 	if (revents == 0) {
2495 		if (events &
2496 		    (POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM |
2497 		     POLLRDBAND)) {
2498 			selrecord(td, &so->so_rcv.sb_sel);
2499 			so->so_rcv.sb_flags |= SB_SEL;
2500 		}
2501 
2502 		if (events & (POLLOUT | POLLWRNORM)) {
2503 			selrecord(td, &so->so_snd.sb_sel);
2504 			so->so_snd.sb_flags |= SB_SEL;
2505 		}
2506 	}
2507 
2508 	SOCKBUF_UNLOCK(&so->so_rcv);
2509 	SOCKBUF_UNLOCK(&so->so_snd);
2510 	return (revents);
2511 }
2512 
2513 int
2514 soo_kqfilter(struct file *fp, struct knote *kn)
2515 {
2516 	struct socket *so = kn->kn_fp->f_data;
2517 	struct sockbuf *sb;
2518 
2519 	switch (kn->kn_filter) {
2520 	case EVFILT_READ:
2521 		if (so->so_options & SO_ACCEPTCONN)
2522 			kn->kn_fop = &solisten_filtops;
2523 		else
2524 			kn->kn_fop = &soread_filtops;
2525 		sb = &so->so_rcv;
2526 		break;
2527 	case EVFILT_WRITE:
2528 		kn->kn_fop = &sowrite_filtops;
2529 		sb = &so->so_snd;
2530 		break;
2531 	default:
2532 		return (EINVAL);
2533 	}
2534 
2535 	SOCKBUF_LOCK(sb);
2536 	knlist_add(&sb->sb_sel.si_note, kn, 1);
2537 	sb->sb_flags |= SB_KNOTE;
2538 	SOCKBUF_UNLOCK(sb);
2539 	return (0);
2540 }
2541 
2542 static void
2543 filt_sordetach(struct knote *kn)
2544 {
2545 	struct socket *so = kn->kn_fp->f_data;
2546 
2547 	SOCKBUF_LOCK(&so->so_rcv);
2548 	knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
2549 	if (knlist_empty(&so->so_rcv.sb_sel.si_note))
2550 		so->so_rcv.sb_flags &= ~SB_KNOTE;
2551 	SOCKBUF_UNLOCK(&so->so_rcv);
2552 }
2553 
2554 /*ARGSUSED*/
2555 static int
2556 filt_soread(struct knote *kn, long hint)
2557 {
2558 	struct socket *so;
2559 
2560 	so = kn->kn_fp->f_data;
2561 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2562 
2563 	kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
2564 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2565 		kn->kn_flags |= EV_EOF;
2566 		kn->kn_fflags = so->so_error;
2567 		return (1);
2568 	} else if (so->so_error)	/* temporary udp error */
2569 		return (1);
2570 	else if (kn->kn_sfflags & NOTE_LOWAT)
2571 		return (kn->kn_data >= kn->kn_sdata);
2572 	else
2573 		return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
2574 }
2575 
2576 static void
2577 filt_sowdetach(struct knote *kn)
2578 {
2579 	struct socket *so = kn->kn_fp->f_data;
2580 
2581 	SOCKBUF_LOCK(&so->so_snd);
2582 	knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
2583 	if (knlist_empty(&so->so_snd.sb_sel.si_note))
2584 		so->so_snd.sb_flags &= ~SB_KNOTE;
2585 	SOCKBUF_UNLOCK(&so->so_snd);
2586 }
2587 
2588 /*ARGSUSED*/
2589 static int
2590 filt_sowrite(struct knote *kn, long hint)
2591 {
2592 	struct socket *so;
2593 
2594 	so = kn->kn_fp->f_data;
2595 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
2596 	kn->kn_data = sbspace(&so->so_snd);
2597 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
2598 		kn->kn_flags |= EV_EOF;
2599 		kn->kn_fflags = so->so_error;
2600 		return (1);
2601 	} else if (so->so_error)	/* temporary udp error */
2602 		return (1);
2603 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2604 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2605 		return (0);
2606 	else if (kn->kn_sfflags & NOTE_LOWAT)
2607 		return (kn->kn_data >= kn->kn_sdata);
2608 	else
2609 		return (kn->kn_data >= so->so_snd.sb_lowat);
2610 }
2611 
2612 /*ARGSUSED*/
2613 static int
2614 filt_solisten(struct knote *kn, long hint)
2615 {
2616 	struct socket *so = kn->kn_fp->f_data;
2617 
2618 	kn->kn_data = so->so_qlen;
2619 	return (! TAILQ_EMPTY(&so->so_comp));
2620 }
2621 
2622 int
2623 socheckuid(struct socket *so, uid_t uid)
2624 {
2625 
2626 	if (so == NULL)
2627 		return (EPERM);
2628 	if (so->so_cred->cr_uid != uid)
2629 		return (EPERM);
2630 	return (0);
2631 }
2632 
2633 static int
2634 somaxconn_sysctl(SYSCTL_HANDLER_ARGS)
2635 {
2636 	int error;
2637 	int val;
2638 
2639 	val = somaxconn;
2640 	error = sysctl_handle_int(oidp, &val, sizeof(int), req);
2641 	if (error || !req->newptr )
2642 		return (error);
2643 
2644 	if (val < 1 || val > USHRT_MAX)
2645 		return (EINVAL);
2646 
2647 	somaxconn = val;
2648 	return (0);
2649 }
2650